Hi all, I am new to xgboost and I am using it to learn two variables (a
and b
) using a 10 x 10 million data set. The parameters that I am using are
params = {‘tree_method’: ‘gpu_hist’,
‘max_depth’:11,
‘learning_rate’:0.05,
‘subsample’:0.80,
‘reg_alpha’:10.,
‘reg_lambda’:10.
}
and i train with 100 rounds.
Basically, the reconstruction looks like it didn’t pick up the signal. Y axis is the reconstructed a
. X axis is the truth. the red line is just a diagonal.
My conclusion would be that the classifier is not complex enough. However, when i increase the max_depth the model performs much worse, or even go to the degree where i stop having any sensitivity at all.
Can anyone give me a hint on what to do?